Grab A Piece Of Big Blue's Big Iron
Alowishus writes: "IBM is going to make one of its high-end zServer mainframes available for free to the public for development and testing of Linux applications. It has 10 CPUs and 2.1TB of storage, and will offer TurboLinux or SuSE distributions set up as virtual servers. It's expected to support about 1,000 simultaneous users." However, hold your horses just a bit: Registration is not yet open, the accounts are good only for a limited time, and the site lists other conditions details, though none sound onerous. Among other things, "once a user is registered and approved to access a LCDS system, a user is required to have direct Internet connection, via a Telnet and SSH client." Though there have been other free sandbox accounts, having an account on an S/390 would be sweet, eh?
I think this is IBM's argument: Linux and Apache have become de facto multi-vendor marketplace standards. It's more effective to adopt and adapt this giant body of work than it is to bear all the costs of a specialized big-iron operating system and web server.
From the Free Software point of view: a quick empirical view of the world shows that corps and govs want big iron, and want some software to run on it. I believe that the world is a better place when all software is available under some kind of open license, not just the bitty box software that you care about. What if I wanna teach myself how big servers run 1000 separate system images? I start reading the arch/s390 directory in the linux kernel source. Or I read a book by a teacher who has access to arch/s390 source and can use that knowledge to teach me.
Computer programming is more than those 10-week class projects in undergrad school. We have millions of lines of "prior art". The GPL license and the BSD license are the most effective ways on the planet to make that prior art as accessible as the primary literature in disciplines like chemistry. That's important if big-iron programming is going to be an open competitive field, not limited to employee/serfs operating under NDA's. (Insert return-of-medieval-guilds rant here).
(Maybe you think the world would be a better place if nobody had a computer that big. I think that would be a Luddite opinion. The fact is, lots of organizations already do, and I'm happier if they are running Linux rather than a closed-source OS like AIX or Solaris).
No, just those sunglasses that block out harmful blue light..
No no, some people do miss a thing here..
/root /usr, 128MB RAM, disk space etc..
/usr/local and start installing/erasing stuff, you'll have to reconsider other users when you running you wild intense-use-of-processor app..
What IBM will give you is an ENTIRE VIRTUAL MACHINE! - not just a very limited user account - but a full Linux virtual machine - with
On Compaq machine you just get an account - you cannot go to
Hetz (Heunique)
From what I understand, linux running on these mainframes is just like linux on your desktop, except that there are several copies running at once.
So, using this, I just like using any other linux machine, and it closely resembles a typical ppc machine. Why do you need an account on a machine like this that's shared with 1000 other users if you can just see if your app compiles on a ppc machine.
They don't actually say if ibm is going to offer anything special. The big appeal with using a machine like this is the high speed internal networking, being able to run a firewall, 5 web servers, 4 app servers, and 2 databases, all on the same machine, but under different operating systems, or use it for mass virtual computer hosting. I don't think that's going to happen with a free acount.
So what's so special about this?
I uploaded some homebrew benchmarking software to it. The memory bandwidth seems amazing. Compared to a reasonably high-end Intel server, it has about 10x the memory bandwidth. Either that, or lots and lots of L2 cache. But I'm sure I wasn't the only one on the system, and the tests consistently showed almost 10x the throughput of Intel.
The test was simple: take a range of memory of X megs, and write a random char to it. Time how many can be done in a quarter of a second, extrapolate MB/s from that. I get almost the same numbers if I use 10MB chunks as I do to 1MB chunks. Good shit here.
Single CPU performance isn't much different, but that wasn't the point. The systems could also be configured to give my account access to multiple processors.
If I started my own hosting company, I think I'd definitely use one of these babies.
Not likely. Yes...it would lock up YOUR virtual machine. And it would make infinte directories on YOUR virtual machine/disk. But that is about it.
These things are amazing. I played on one in college a lot (well, a 3090). It was running 3 copies of McGills MUSIC system, VM/CMS and a few other things. One MUSIC system would be dirt slow cause it had a ton of students on it taxing the virtual system. Everything else one the 3090 ran just fine.
This is not the same as running multiple processes on a PC where the performance of one affects that of another. This is akin to 300 machines sitting side by side.
The main benefit of SSH is that it sends everything encrypted. If you need to log in and administer your corporate web site there really isn't much choice lest someone tap your line and guble your root password.
However this box is for public access. In real life the users won't have anything even remotely confidential on it. Nothing at least that isn't also available by CVS from public servers (like cvs.kde.org).
Sure telnet may be a bad habit but it isn't a sin.
BTW: This report talks about the free mainframe in a different light from all the others. Before this all you herd of was people getting this box to fine tune mainframe ports of Linux software. Sure you let reporters run lynx and write about it but the main tool on this box is GCC.
--= Isn't it surprising how badly I spell ?
http://www-1.ibm.com/servers/eserver/zseries/os/li nux/lcds/index.html
9672 G6 Model ZX7 (10 way processor)
Memory:
32GB
Storage:
Shark 2105-F20 (2.1 terabytes)
Software:
Z/VM
VM PRF
C Runtime Library
REXX Compiler and Library
VM TCPIP (FL32A with SSL support)
DIRMAINT
Tivoli Storage Manager Version 4.1
VM TCPIP NFS feature
VM EREP
(No clue what most of the above mean)
The minimum memory you can stick in a zSeries is 1 gigabyte. The maximum is 64 gigabytes.
Storage throughput has similarly increased since your employer purchased their dinosaur, I just couldn't find figures for what they are now.
Charles Miller
--
The more I learn about the Internet, the more amazed I am that it works at all.
I can see it now, the ultimate platform for playing CoreWars.
Better start writing your bots now!
For instance, when the <plug>GnuCash</plug>developers get the RPC-enabled, PostGres-driven backend fully up to speed (it works, but it's not production-use material yet) it'd be a blast to port it to this architecture. If we did, do you really think IBM would cut off access?
Go you big red fire engine!
Any sufficiently advanced technology is indistinguishable from a rigged demo
--Andy Finkel (J. Klass?)
Why code in C when I can code in COBOL
:)
OS/390 has a very high quality C compiler. I've been using it for the last two years
Finkployd
Stick with it! I'm doing the same as you right out of college and having a blast. Mainframes aren't going anywhere (sales are up 40% last I heard) and most of the systems programmers are close to retiring. We are going to be in high demand, my friend :)
Finkployd
Actually I would be suprised of OS/390 were running anywhere on that machine. Most likely it's running VIF (Virtual Imagining Facility, a stripped down version of the old VM) to control the linux images.
Finkployd
Thanks you the link, you are quite correct.
I'm a bit suprised they are not using this oppertunity to show off the new VIF product.
Finkployd
Great! Let's all just shout foul to IBM. How dare they give access to a computer for free? Bad bad IBM! It should be GPL! No! They should also give free T-shirts!
SourceForge is a great contribution but don't use it to put down other gifts. That's greed.
You can only use up the processor time that is allocated to your partition. The other users wouldn't even notice the difference if you used 100% of your CPU allocation. The whole VM can be admin'd quite closely, so it isn't hard to keep things from getting out of hand if you set things up properly from the start.
--
"It's tough to be bilingual when you get hit in the head."
Right, it is essentially a PC server on a PCI card, with access to all of the I/O advantages that goes along with being in an AS. Provides nice backup and central administration capabilities, and you can reload the Windows image quickly if your app hoses things up, and still have all of your user data nicely separate.
Linux is running on the mainframe as much as OS/390 or z/OS runs on the mainframe... everyone runs on top of a VM on those machines.
--
"It's tough to be bilingual when you get hit in the head."
We don't like Microsoft, but we (in general) like IBM, so we'll do IBM a few favours if it benefits us too.
------
What about Progeny? It's got a flashy GUI and it's Debian-based-and-compatible.
------
Setting up a set of VMs inside the VM would be the really fun part of having this whole system.
------
It would be nice if database developers could get access to database space on the latest from a database company, thus helping the database company get the latest features to be used by developers. I do work for Oracle so I would hope Oracle would pick up this idea first.
Be Free: Free Software Tuition
This is great marketing for both IBM, and Linux. First off I think this is more of a marketing tactic than IBM trying to be helpful to any developer. By having thousands of developers do their thing on the machine, they could always turn around and pimp the results as both an IBM, and joint Linux effort which is pretty cool for Linux marketing...
However on the flip side of the coin, I hope the developers rush to fill these slots as opposed to some troll who's going to use those accounts for silly shit like h4x0rf00.c programs they wanna throw up, or uneccessary other shit...
Let's kill some Americans and blame Cuba
Want Root?
Well, any developer worth their salt knows how to set up a hotmail account, and use that to re-apply for another period.
Say no to software patents.
... a cluster of thousands of well connected DeCSS/OT VII/<insert your favorite censored item> mirrors? Or will they firewall off HTML access to those virtual linux boxen? Wouldn't it be ironic if mirrors of this stuff turned up as well? IBM may have unwittingly set up a giant community blackboard here...
Say no to software patents.
Think about it. Nobody cares whether this runs on an IBM mainframe, or on a Beowulf clusters of gameboys. However, this thing sits right in the center of IBM's Noc, and is thus probably very well connected. Think Gigabit connections directly to the important backbone providers... If you have something interesting to publish, you put it at that place. You don't care about the OS, nor the hardware: you use it solely for its outstanding connectivity!
Say no to software patents.
AFAIK, it does not run an x86 emulation. It's a port to the mainframe's processor, just like there are ports to the Alpha, the PowerPC, the Sparc, etc. Virtualization only goes as far as is needed to compartimentalize the mainframe into zillions of server, it does not emulate the processor.
So you would only have source compatibility, and no binary compatibility, and thus it does make sense to test your software there, in order to see whether it ports allright to this architecture.
Say no to software patents.
What a waste to spend resources on opening an S/390 considering what they could be used for.
What about taking 10% of that BILLION and earmarking it to support open source developers??
Where were they when Eazel tanked? What about the folks at SourceXchange? Are they doing anything more than thinking about marketing, pr facetime, and beating their own products to death?
I'd much rather see some of that money go to supporting hookup of an IBM microdrive to the Agenda, or a zillion other things, than this.
IBM should earmark 0.05% of their budget (that's still half a million bucks right?) to - guess what - pay great open source based developers and designers to build a site that would try to get feedback from the Linux community, including developers, users, and purchasers, as to what sort of things we'd like to see. It might even save them some marketing money. IBM's done some good things but this is not the top priority if they are serious about spending that money on open source.
If they believe Linux gives them value for the money, then they ought to be willing to put money down to get high quality engineering and design talent to work on projects which IBM could share with the open source community and continue to improve Linux.
One really cool thing they could do is endow a chair (or 10 or 20) like the year off from school which Perl mage Damian Conway received from the community.
isn't that one of the cool things about linux, that you don't NEED an expensive machine to run it? sure, for deploying something that demands a lot of power, one of these expensive servers would be great, but why put 1000 developers on it? i'm sure anyone with the skills to develop a linux application has access to some old pc to run it on (and you need SOME box in front of you in order to ssh to ibm's server anyway).
this just sounds like marketing hype, and not nearly as cool hype as the spraypainting thing...
But what might seem like a grand experiment is also a shrewd marketing move by IBM. None of IBM's server competitors--such as Sun Microsystems, Hewlett-Packard or Compaq Computer--has offered similar programs.
ZDnet seem to be avoiding the fact that Compaq have their huge TestDrive program. They let anybody (currently)access:
Beowulf Cluster on AlphaServers
Caldera OpenLinux on ProLiants
Tru64 on various AlphaServers
TruCluster server on several AlphaServers
OpenVMS on AlphaServer
Debian on AlphaServer
Debian on ProLiant
FreeBSD on AlphaServer
FreeBSD on ProLiant
Kondara Linux 2000 on AlphaServer
Kondara Linux 2000 on ProLiant
Linux64 on Blazer Itanium
Mandrake on Proliant
NetBSD on on AlphaServer
RedHat on many ProLiants
RedHat on even more AlphaServers
SuSe on ProLiants
SuSo on AlphaServers
TurboLinux on ProLiant
Plus numerous databases...
Lots of toys... all for free...
I agree that the limited time is good for the 'testing' users who just want to know how it feels, but I doubt they will appeal to someone who would like to do a serious work with such conditions.
If programs would be read like poetry, most programmers would be Vogons.
You know, I know that IBM is going to donate/spend $1 billion this year on Linux and open source, and this is clearly part of the whole plan. And while I would be the very last to look a "gift horse in the mouth" so to speak, wouldn't it be even more super if they dumped some money into these ailing open source projects or hired developers a.k.a. VA Linux?
What will you need if you want to try Windows out? A Beowulf of Crays?
http://www-1.ibm.com/servers/eserver/zseries/os/li nux/lcds/index.html
SGI Origin 3X00 series. Up to 512 MIPS R14000/500 CPUs, 1 TB RAM. ~700 GB/sec system bandwidth. Add as many PCI or XIO expansion chassis as you desire. (VME can be added via XIO). Add up to 16 "G-Brick" Onyx 3000 / InfiniteReality 3 graphics subsystems. Add up to 36 individual "snowball" DM3 dual channel SD/HD digital video cards for editing HDTV. Add up to 768 SCSI and/or Fibre Channel XIO cards.
Be sure to win the lottery. Err... Be sure to win several lotteries.
http://www.sgi.com/origin/3000/3800.html
http://www.sara.nl/Customer/systems/sgi3800/
No Linux yet, just IRIX + some Cray libraries (SCSL). Can be partitioned if you want to run different instances of IRIX. Can also be clustered via ethernet, gigE, HiPPI, or GSN (800 MegaBytes/sec HiPPI derrivative).
I hate to do it, but I have to wonder what IBM's *real* intentions are. Also, just who will get accounts on this machine? The big names: Torvalds, Cox, ESR, RMS? Friends/customers of IBM? The real coders (Apache members, GNOME members, etc)?
I'm almost scared that they may just use this as a way to collect information about up-and-coming projects. Even if I'm working on a GPL'ed project, I'm somewhat weary of compiling and running it on a machine owned by someone else. One day it may be my app... the next it could be "IBM MEGA COMMERCIAL APP". Maybe not the code, but perhaps the idea. I can't compete with IBM when it comes to resources.
Buyer Beware.
Are you 5 years old? Grow up already. {buhahuh he said load, then he said tool}
Enigma
Enigma
One word: seti@home
Many years ago, Digital did this with their Alphas when they first came out around 1990. They did everything they could to bring attention to these fast guys, including putting out a number of white papers detailing its architecture and core design. Somehow they still could not break the Intel barrier despite their speed.
They gave logins to anyone who asked for one in order to see what could be done with the systems. They were always overloaded and it seemed like there was great interest in the machines, but eventually $$$ and non-native Intel compatibility limited them. Good luck to IBM.
"Beware of he who would deny you access to information, for in his heart, he dreams himself your master."
Well it DOES.
--------
Bleah! Heh heh heh... BLEAH BLEAH!!! Ha ha ha ha...
A Beowulf cluster of these would be quite silly. The main selling point of the z900 is that it is massively partitionable. It puts the Compaq/Digital GS320 and Sun Enterprise 10000 to shame. If you're running just one application or computation, partitioning doesn't serve a purpose.
The z900 is designed for insane availability (It's relative, the G3, runs the US air traffic control system). Features, such as partitioning and the impressive Dynamic CPU sparing ensure that the computer is always running at full capacity. These features are not necessary for scientific clusters and defeat the purpose of Beowulf's impressive price/performance ratio.
Please read the text carefully. IBM makes this available for developers to port and test their software on IBM mainframes + Linux combination. This will reduce bugs for IBM and they can sell more of these machines especially to busineeses. It looks like having the ability to run linux on this machines is very valuable for businesses and IBM want more software to be compatible with their machines.
Will this be like open bsd's jail? Or will you be able to use it like its your own colo system? What potential do most slashdoters have for this?
The Lottery:
"Not my manner of thinking but the manner of thinking of others has been the source of my unhappiness." - M
Sure it would be hard to fill it up via the net, but it wouldnt take long to generate a large file *on* the system.
The Lottery:
"Not my manner of thinking but the manner of thinking of others has been the source of my unhappiness." - M
Are you trolling? At that point, IBM would be investing in possibly competing firms which do not necessarily even have a snowball's chance in hell of being successful. And it would be showing favoritism to an outside group of developers. Should they also give money to RedHat? Would SourceXchange survived (if they indeed are dead) if it had some IBM money propping up their carcass? There is still some Darwinism inherent in the Open-source model.
IBM should earmark 0.05% of their budget (that's still half a million bucks right?) to - guess what - pay great open source based developers and designers to build a site that would try to get feedback from the Linux community, including developers, users, and purchasers, as to what sort of things we'd like to see.
I heard that's what SourceForge and even Slashdot were about. Before you find that odd, both sites are actually very good barometers of peoples' desires. And plus, they already have alphaWorks and developerWorks, both of which I visit on a regular basis.
Please don't get me wrong, I too am saddened and disappointed at Eazel's demise. But I appreciate the fact that IBM has adopted a mostly hands-off approach to third party open source development.
Consider that IBM did decide to fund a company like Eazel, what makes that any better than Microsoft funding Corel? I think IBM is doing the "right thing", and avoiding lots of nasty accusations in the process.
Yeah, all the specs for the machine are there (10 CPUs, 32GB RAM) and 2.1 TB disk and they still call it DASD! I'm having bad flashbacks to allocating cylinders and other IBM doublespeak.
-----------------
www.lucernesys.comHorizon: Calendar-based personal finance
I expect that the majority of work allowed by Big Blue will be architecture porting and testing (eg, x86 to S390), rather than development from scratch. There of course will be (I expect) some products that may wish to exploit the Linux-to-z/OS (the OS formerly known as OS/390, MVS, yadda-yadda) functionality that is soon to come (why go over the network from a logical Linux machine to a logical z/OS machine when you can go cross-memory?), and these will be developing some code from the gound up. But I can see that a lot of software currently running successfully on x86, PPC, etc (commercial or not) needs to be tested on 390 to officialy "support" it, and, let's face it, not many people have easy access to a 390 machine, let alone a 390 machine running Linux!
:-)
This is IBM's way of getting as much existing Linux software as possible to list 390 as a supported architecture. There's a lot of support-contract related money to be made by distributions in this areana....if a company has already shelled out on 390 hardware, they are hardly going to go without a software support contract for their Linux Distro.
Having said all that, I believe that the porting effort is negligable for most user-level applications, but of course, you would like someone to actually test their software on the architecture before assuring you it works there, wouldn't you?
Gollo.
A number of posts have discussed IBM's intentions and whether or not this is just a hype device. I just wanted to chime in and mention that I have been part of a Linux startup for over 2 years now and IBM have been incredibly responsive to any input or criticisms our group has offered. I know that IBM views Linux as their opportunity to gain ground in a server market that they have lost to Sun largely because of the OS and less because of the hardware. IBM would like to encourage the growth of the 390 largely because they feel it is fundamentally a superior product, but is losing ground because of the application space. To that end (and I was actually present at a meeting w/ a General Manager and 3 VP's) when someone mentioned the idea of making time available. Everyone's reaction was enthusiastic. For them, here was a great opportunity to be viewed by the best the open source communitiy has to offer and allow IBM the opportunity to prove that they do in fact offer superior hardware. Things like limited time I imagine are only there to prevent squatters from wasting people's time. Beyond that, IBM seems to have demonstrated a complete commitment to Linux on all levels, so this idea really just seems to be the next obvious evolution of that commitment. I do have to admit though, the spraypaint thing was cooler IMHO. simple4